23
Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

  • View
    221

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Old Dog Consulting

Multi-Segment Pseudowires: Recognising the Layer Network

Adrian Farrel

Old Dog Consulting

Page 2: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 2© Copyright Old Dog Consulting 2010

Agenda• Existing building blocks

• Protocol layering in the data plane

• Multi-segment pseudowires• Architecture and drivers

• Functional requirements• Picking paths and setting up pseudowires• Service-level requirements

• The layer model• Pitfalls and benefits• Next steps

Page 3: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 3© Copyright Old Dog Consulting 2010

The PW Layering Model

• RFC 3985 defines logical protocol layering

• For example…

Payload

Encapsulation

PW Demultiplexer

PSN Convergence

PSN

Data-Link

Physical

May be empty

May be empty

Ethernet Header

MPLS Tunnel Label

MPLS PW Label

Control Word

IP Header Data

First byte Last byte

Page 4: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 4© Copyright Old Dog Consulting 2010

Multi-Segment ArchitecturePseudowire Segments

Provider Network

TunnelSwitching PEs

Native Service ACTerminating PE

Provider Network

• Simple extension to the RFC 3985 model• Emulated service is still CE-to-CE• Tunnels are still used to carry the PWs• End-to-end PW is called a multi-segment PW

• Runs between the Terminating PEs (T-PEs)• Constructed from PW segments

• Carried across provider networks in tunnels• Tunnels terminated at PEs• PW segments “switched” (or stitched) at Switching PEs (S-PEs)

CE

Page 5: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 5© Copyright Old Dog Consulting 2010

MS-PW Deployment Motivations• Initial model shows inter-AS PW service• A more pressing need

• Reduce the complexity of the tunnel mesh• Help scaling at PEs and P nodes

• S-PE becomes a network-internal node• Not the best name!

• Same model applies to inter-area PW service• Utility extends to P2MP PWs (discussed later)

Pseudowire Segments

Provider Network

Tunnel

Switching PE

Native Service ACTerminating PE

CE

Page 6: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 6© Copyright Old Dog Consulting 2010

MS-PW Challenges• Data plane encapsulation• Picking a path through the network• Setting up pseudowire segments and PSN

tunnels• Service-level requirements

• Capacity• Diversity• P2MP

• Operations, Administration and Maintenance

Page 7: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 7© Copyright Old Dog Consulting 2010

Data Plane Challenges• PW encoding should be independent of PSN technology

• Same techniques/hardware “packetization”• Regardless of underlying PSN transport

• Resource reservation is needed to guarantee PW service• PWs use PSN tunnels

• Reservation must use tunnel resources• Tunnel must map its resources to network resources

• Tunnel transit nodes are not aware of payload PWs

• PWs must be multiplexed onto data channels to scale the data plane

• PW flows must not merge• Have to be able to trace and distinguish individual PWs• Essential for OAM and fault diagnosis

Page 8: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 8© Copyright Old Dog Consulting 2010

Path Determination• Choices to be made

• Which tunnels to use?• Which S-PEs to use?• For dual-homed CEs: which T-PEs to use?

• Are these choices made in planning or during LSP set-up? (see next slide)

• What factors affect the choices?• Tunnel load and capacity• S-PE load and capacity• Reduce the number of segments on the path?• Path diversity for backup services

Page 9: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 9© Copyright Old Dog Consulting 2010

PW Setup• Components of MS-PW establishment

• Tunnel set-up• PW segment set-up• PW segment stitching

• Choices• All through the management plane• Control plane for tunnels and MP for PWs• Control plane for tunnels and PW segments

• But segment stitching using management plane• Fully in the control plane• Some segments MP, some segments CP

• How much operator involvement is needed?• Where are the administrative boundaries?• Can the signalling protocols handle the MS-PW path and

constraints?

Page 10: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 10© Copyright Old Dog Consulting 2010

Service Requirements• Influence path determination and set-up• PW capacity and quality requirements• Protection considerations

• End-to-end protection• Tunnels are diverse• No re-use of S-PEs

• Segment protection• Tunnels between S-PEs are diverse• Protect a PW segment• Protect an S-PE

• Point-to-multipoint PWs• Use a single P2MP tunnel?• Stitch multiple P2P PW segments?• Combine the techniques?

Page 11: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 11© Copyright Old Dog Consulting 2010

MS-PW Protection

End-to-end MS-PW protection

Single segment protection

Multi-segment protection

PSN tunnel protection

Page 12: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 12© Copyright Old Dog Consulting 2010

P2MP Pseudowires

Page 13: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 13© Copyright Old Dog Consulting 2010

OAM Challenges• OAM function provides

• Service verification• Fault detection and reporting• Fault isolation

• Service verification is end-to-end• Can run OAM on the PW or on the emulated service

• Faults need to be known where they are to be handled• T-PEs for end-to-end protection• S-PEs for protecting individual segments

• Scaling may be an issue• How many PWs pass through an S-PE?• Running OAM on a tunnel can solve this

Page 14: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 14© Copyright Old Dog Consulting 2010

The Layer Model

• There is a natural layering available• Nothing clever!

• Make a topology of• Nodes = PEs (T-PEs and S-PEs)• Links = PSN tunnels

• See that these links have cost and bandwidth

• Plan and set up MS-PWs on this topology• Each “hop” is a single segment PW

Page 15: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 15© Copyright Old Dog Consulting 2010

Topology Layering

• Tunnels between S-PEs in the PSN become links in the MS-PW network

Page 16: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 16© Copyright Old Dog Consulting 2010

Multi-Access links

• P2MP tunnels form multi-access links in the MS-PW network• Care needed about unidirectional P2MP tunnels!

Page 17: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 17© Copyright Old Dog Consulting 2010

The Application Layer is Extra• Emulated service is between CEs

• CE is out of scope for the provider network• End-to-end protected service is required

• Protected service is through two “parallel” emulated services• Individually requested• Different T-PEs?• Different ACs?

• The protection is the responsibility of the service user• But the emulated services need to have disjoint paths

• Requires the use of Shared Risk Link Groups (SRLGs)

Page 18: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 18© Copyright Old Dog Consulting 2010

PWs Are Transport-Agnostic

• No surprises here• But packet technologies can be different• Architecture must allow independence of PW segments• Still deliver end-to-end emulated service

MPLS-TP IPEthernet

T-PE S-PE

CE

AC

Packet Tunnel

PW Segment

End-to-End PW

Emulated Service

Stitch

Switch

Page 19: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 19© Copyright Old Dog Consulting 2010

Pitfalls to Avoid

• “Don’t worry about the control plane”• Let’s do it all in the management plane for now

• True, but network planning can make good use of layers• OAM layering will help operations

• “Let’s leverage the IGP”• We can use our IP/MPLS IGP “discover” S-PEs

• Fine to run an IGP instance in the PW layer• But don’t overload the normal IGP• Consider how inter-AS will work

• How about using PCE in the PW layer?

• “Layers add unnecessary complexity”• We only have a simple network with one S-PE

• Networks will inevitably get more complicated and larger• How easy will it be to cut over to a layered approach later?

Page 20: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 20© Copyright Old Dog Consulting 2010

More Pitfalls• “Network layering implies operational separation”

• We want to operate an integrated PSN• Network layers can be operated and planned independently• Dynamic integrated multi-layer networks are possible

• Feedback loops between layers with appropriate policy controls and operator input

• IP/Optical is the latest buzz in this area

• “We can grow LDP to handle MS-PW” in IP/MPLS networks• We already use LDP for PW set-up

• It’s true, any protocol can be extended to do anything!• LDP is designed as a neighbour-to-neighbour protocol• T-LDP is currently used only for single segment PWs• Functional creep does not make for good protocol design• Need extensions for all elements of constraint-based path signalling

• Explicit routes, route recording, bandwidth reservation, protection, path association and path diversity, etc., etc.

• We do already have a PSN signalling suite for this type of function (RSVP-TE/GMPLS)

Page 21: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 21© Copyright Old Dog Consulting 2010

Potential Benefits?• Simplified network view

• Aids operation and planning• Integration of multiple PSN types• Reduced complexity in network operation

• Separation of application from operation• Reduced number of control plane protocols

• Increased features and functions• Leverage experience with existing multi-layer

networks and control planes• Path computation and control • Resource reservation and management• Protection and restoration• Point-to-multipoint• OAM control and configuration

Page 22: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 22© Copyright Old Dog Consulting 2010

What Should We Do About It?

• Decide whether MS-PWs are for real

• Plan our control plane protocols• Don’t just evolve them piecemeal• Look to see if we can leverage existing

protocols we are already running

• Recognise the layered architecture• Build this into our PW architecture work• Design PW networks as layers

Page 23: Old Dog Consulting Multi-Segment Pseudowires: Recognising the Layer Network Adrian Farrel Old Dog Consulting

Page 23© Copyright Old Dog Consulting 2010

Questions

[email protected]